A vendor called Kushal Sharma to tell him his $18,000 contract was now a $28,000 contract. That is normally the part of the story where the ops leader begrudgingly signs. He didn't have to — and the reason why had nothing to do with negotiation.

Kushal leads the internal AI organization at Circle, a function he got to build because he had already spent two years merging Circle's data and revenue operations teams under one mandate: turn the ops people into product owners, turn the analysts into engineers, and put real architecture underneath the go-to-market motion. Medallion layers. A dbt semantic layer. Orchestration he controlled. Lineage on everything. Before Circle, he spent eight and a half years at Hootsuite, where he led the company's data organization.

Then AI actually arrived — skills and agents shipped — and he was one of the very few operators in the market with something for them to operate on. Because this one goes deeper than usual, LeanScale CTO Jake Toepel joins Anthony Enrico to go layer by layer: what a semantic layer actually is, what a context graph actually does, when a vector database is worth buying, and why almost none of this is an LLM.

WHAT YOU'LL LEARN

How owning your orchestration layer quietly removes a vendor's pricing leverage before they ever call

What a context graph does, why RAG existed before it, and why an LLM is not a store of knowledge

How Circle built an AI SDR that generated close to seven figures, and what actually made it work

CHAPTERS

00:00 — Cold open + intro

02:11 — The vendor call: an $18K contract becomes $28K

05:07 — Owning the data layer means owning your leverage

06:50 — What a semantic layer actually is

12:48 — The litmus test — if a human can't read it, neither can your AI

19:47 — Medallion architecture: bronze, silver, gold

26:17 — Context graphs from first principles

35:19 — "None of this is an LLM" — the software underneath

40:12 — The maturity ladder: when to actually buy a vector database

49:09 — Stop throwing your best model at a simple problem

51:33 — What it unlocks: a Slack bot in 10 minutes, and an AI SDR

1:04:27 — Relevance or support center: the choice facing RevOps

1:07:09 — Wrap

ABOUT THE GUEST

Kushal Sharma leads the internal AI organization at Circle, the community platform used by more than 15,000 communities. He joined Circle in 2024 to run a combined data and revenue operations function and spent two years putting real architecture underneath the go-to-market motion: medallion layers, a dbt semantic layer, orchestration the team controls outright, and lineage on everything. That foundation is what let Circle stand up an AI function when agents and skills actually shipped. Before Circle, Kushal spent eight and a half years at Hootsuite, where he led the company's data organization.

ABOUT THE LEANSCALE PODCAST

The LeanScale Podcast is the show for GTM operators building the next generation of revenue infrastructure. Hosted by Anthony Enrico, co-founder of LeanScale, we go deep with the operators, founders, and executives shaping how modern go-to-market teams are designed, scaled, and transformed.

▶ Subscribe for new episodes

▶ Learn more about LeanScale: https://www.leanscale.team

▶ Follow Anthony on LinkedIn

▶ Follow Jake Toepel on LinkedIn

▶ Follow Kushal on LinkedIn

KEYWORDS / TAGS

RevOps, Revenue Operations, GTM, GTM Engineering, AI in RevOps, AI Agents, AI Infrastructure, Semantic Layer, Context Graph, Vector Database, RAG, Medallion Architecture, Data Lineage, Data Warehouse, dbt, Analytics Engineering, Data Orchestration, LLM, Model Selection, AI SDR, Sales Operations, Salesforce, Snowflake, Circle, Hootsuite, B2B SaaS, RevOps Leadership, Data Org, LeanScale, Anthony Enrico, Jake Toepel, Kushal Sharma

HASHTAGS

#RevOps #GTM #AI #AIAgents #RevenueOperations #DataEngineering #AnalyticsEngineering #SemanticLayer #ContextGraph #VectorDatabase #dbt #B2BSaaS #AITransformation #LeanScale #LeanScalePodcast

The Agent Was Never the Hard Part | Kushal Sharma on the Data Layer Underneath AI

The LeanScale Podcast with Anthony Enrico

The Agent Was Never the Hard Part | Kushal Sharma on the Data Layer Underneath AI